Tube Bridge MCP Server

tube-bridge

Self-hosted YouTube MCP server for AI agents.

Search YouTube, inspect videos and channels, fetch transcripts and comments, extract timestamped frames, and build local semantic-search corpora.

Quick start

PyPI

pip install tube-bridge

tube-bridge          # stdio transport
tube-bridge --http   # HTTP transport on port 8080

Docker

docker run --rm -p 8080:8080 ghcr.io/thewhitewater/tube-bridge:latest

The HTTP MCP endpoint is http://localhost:8080/mcp.

Official MCP Registry

Registry name: io.github.TheWhiteWater/tube-bridge

Registry-aware clients can install the PyPI distribution with uvx and launch the stdio server without a hosted intermediary.

MCP client configuration

Local stdio

{
  "mcpServers": {
    "tube-bridge": {
      "command": "tube-bridge"
    }
  }
}

If the executable is not on your client's PATH, use the full path returned by which tube-bridge.

Streamable HTTP

{
  "mcpServers": {
    "tube-bridge": {
      "type": "http",
      "url": "http://localhost:8080/mcp"
    }
  }
}

Protected HTTP

Set a server-side Bearer key:

export TUBE_BRIDGE_AUTH_KEY="choose-a-long-random-value"
tube-bridge --http

Then configure the client header:

{
  "mcpServers": {
    "tube-bridge": {
      "type": "http",
      "url": "https://your-host.example/mcp",
      "headers": {
        "Authorization": "Bearer <your-key>"
      }
    }
  }
}

/health remains public. /mcp, /sse, and /messages require the Bearer key when TUBE_BRIDGE_AUTH_KEY is set.

Tools

Fourteen tools can run without a YouTube Data API key. Comments, channel search, and channel details require one.

Optional YouTube Data API key

Create a YouTube Data API v3 key in Google Cloud Console, then set:

export YOUTUBE_API_KEY="your-key"

With a key, search, video information, and trending use Data API v3 first. Supported operations fall back to yt-dlp when the key is absent or quota is exhausted.

Local semantic corpus

Corpus data and embeddings stay on the machine running tube-bridge.

corpus_create("ai-agents", "AI Agents Research")
corpus_add("ai-agents", "https://www.youtube.com/watch?v=VIDEO_ID")
corpus_search("ai-agents", "memory systems")
corpus_list()
corpus_delete("ai-agents")

• storage: SQLite with sqlite-vec;
• embeddings: BGE-small-en-v1.5 through fastembed;
• chunking: 80-second windows with 20-second overlap;
• results: similarity score, source time span, video title, and timestamp URL;
• default data directory: ~/.tube_bridge.

Set TUBE_BRIDGE_CACHE to use another directory:

export TUBE_BRIDGE_CACHE="/path/to/tube-bridge-data"

The embedding model may be downloaded on first use.

Frame extraction

youtube_get_frame needs ffmpeg on PATH. The Docker image includes it; source and PyPI users install it with their OS package manager.

Each call downloads only a short temporary section, returns one bounded JPEG, and removes the temporary media before returning.

Proxy support

If YouTube blocks requests from your network, configure an HTTP(S) proxy:

export TUBE_BRIDGE_PROXY="http://proxy.example:8080"

The value is used by yt-dlp and transcript requests. Keep proxy credentials in environment variables, never in MCP configuration committed to source control.

Transports

• stdio — recommended for local MCP clients;
• /mcp — Streamable HTTP;
• /sse and /messages — legacy SSE compatibility;
• /health — process health and tool count.

Agent Plugin preview

GitHub Releases also include tube-bridge-agent-plugin-<version>.zip. It bundles:

• the local stdio MCP configuration;
• the tube-bridge-research skill;
• research templates and source-evaluation guidance.

The plugin format does not install system or Python dependencies. Install Python 3.12+, ffmpeg, and tube-bridge dependencies in the environment used by your plugin host.

Development

git clone https://github.com/TheWhiteWater/tube-bridge.git
cd tube-bridge
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements-release.txt
pip install --no-deps -e .
pip install pytest pytest-asyncio pytest-mock build twine
python -m pytest tests -q

Optional live smoke test:

python test_tools.py

The deterministic test suite does not call YouTube. The live smoke test does.

Known limitations

• YouTube can restrict anonymous yt-dlp and transcript requests, especially from cloud-hosting IP ranges.
• A Data API key improves search and metadata reliability but does not replace transcript access.
• Initial local embedding-model setup may require network access and additional disk space.
• This project is self-hosted software; it does not provide accounts, hosted storage, or a managed endpoint.

Contributing

See CONTRIBUTING.md. Security reports should follow SECURITY.md.

License

MIT — see LICENSE.